Operations Research in Health Care:

Perspectives from an engineer, with examples from emergency medicine and cancer therapy

Timothy Chan
University of Toronto

Steven Brooks
St. Michaels Hospital

Clinical and Population Research Rounds St. Michaels Hospital October 20, 2011

Overview
Introduction to Operations Research (OR) The landscape of OR and health care: three categories of problems
AED location example (policy) Radiation therapy example (micro) Clinic scheduling example (macro)

What is Operations Research?

The most important field youve never heard of A liberal education in a technological world
Boston Globe article, 2004

OR is the study of improving operations and decisions through the use of quantitative techniques
Optimization, probability, statistics, computer modeling, simulation, queuing, game theory, etc. Useful applied math

The lack of a universal definition is both a strength and a weakness of our field
3

Where is OR used?
Historically, OR has been applied to areas such as:
Military: What is the optimal size of a convoy (during WWII)? Manufacturing: How to minimize cost of production while meeting demand? Transportation: How to configure a supply chain to minimize transportation costs? Finance: What is the best basket of stocks to hold to maximize return while keeping risk at an acceptable level?

More recently, OR has been used in the realms of:

Entertainment: Queue management at Disney World Pricing: Dynamic pricing of airline tickets and hotel rooms Sports: Scheduling a season of MLB games Health care:
4

OR applications in Health Care Three categories

Policy
For the system

Cost effectiveness Guidelines in public health

Micro
For the patient

Macro
For the provider

Medical decision making Treatment design

Resource allocation Utilization, throughput

5

A Policy Problem: Public AED location

Collaboration with Steve Brooks, Laurie Morrison Automated External Defibrillators (AED) in hotspots
Determine new hotspots where AEDs should be added or redeployed from colder areas

Automated external defibrillators (AEDs) can be used by bystanders to diagnose and treat a cardiac arrest victim prior to EMS arrival
AEDs useful if nearby

Public access defibrillation (PAD) programs

Public access defibrillation: place AEDs in public locations so that they may be used to treat cardiac arrest victims by lay responders Organizations produce guidelines to help inform public AED deployment
Locations with high historical incidence

Where public AEDs are actually placed can be highly variable from city to city
Specific donors/campaigns may want to see AEDs placed in certain locations
7

How to optimize PAD programs

1. Place AEDs in appropriate geographical locations throughout the city 2. Place AEDs in appropriate buildings and locations within buildings 3. Aid lay responders in finding a nearby AED 4. Make sure AEDs are accessible 5. Ensure responders are willing to operate an AED in an emergency situation

Build awareness

Goals
To develop a methodology that can identify cardiac arrest hotspots in any city and prioritize geographies for AED deployment To identify hotspots that may be missed by other methods To test and validate methodology using data from Toronto

Optimization model
Maximal covering location model
Maximize # of cardiac arrests covered (within certain radius) by deploying AEDs to N locations

10

Data
Cardiac arrests (1310)
Resuscitation Outcomes Consortium Epistry database Location and other info for cardiac arrest cases from December 2005 July 2010 Inclusion criteria: Toronto, public locations, atraumatic, EMS-attended

Currently deployed AEDs (1669)

Registry from Toronto Emergency Medical Services with location info Registration not mandatory; likely more AEDs out there but no visibility by EMS

Potential AED locations (25,851)

Building database from City of Toronto Employment Survey

11

Results

12

Results

13

Results

14

Results

15

Results
Performance metric Total cardiac arrests covered Average distance from cardiac arrest to closest AED Baseline 304 (23%) Optimization N=10 N=20 356 (27%) 273 m 386 (29%) 266 m N=30 416 (32%) 262 m

281 +/- 229 m

16

Observations
Lots of cardiac arrests occur outside
Hard to classify the building type in which they occur, so most analyses miss these locations

Some downtown hotspots have >5 historical cardiac arrests

In more troubled areas Many candidate locations for AED deployment

Hit diminishing returns relatively quickly

Need to balance impact of location optimization with other initiatives that may improve PAD programs

17

Collaborating with Engineers

Provides a fresh viewpoint on a problem Alternative approaches, frameworks, methods may become apparent
Approaches to non-health care problems may lend well to health care problems through analogy

Engineers have tools, methods and expertise not available to the health care researcher

Collaborating with Engineers

Caveats
Engineers can be strange and threatening We speak different languages
Need to invest in mutual education about the art of the other

Spend a lot of time learning about the methods being used and make sure that they make sense for the real world problem
Pay attention to model assumptions and parameters as they relate to the real world problem

A Policy Problem: Summary

Developed mathematical model to optimize locations of AEDs to cover as many (historical) cardiac arrests as possible Cardiac arrest coverage can be improved significantly with a small number of AEDs deployed in priority locations
Diminishing returns reached quickly

Optimization is only a small part of improving survival from OHCA through increased AED use
Any advance that increases the coverage radius can significantly improve performance of the system
20

OR applications in Health Care Three categories

Policy
For the system

Cost effectiveness Guidelines in public health

Micro
For the patient

Macro
For the provider

Medical decision making Treatment design

Resource allocation Utilization, throughput

21

A Micro Problem: Radiation therapy

Optimization algorithms used to design radiation therapy treatments (beam angles, intensity of each radiation beamlet, etc.) Uncertainty (e.g., setup error, breathing motion) can reduce treatment effectiveness Goal: Design treatment plans that are insensitive to uncertainty while achieving other objectives

22

Radiation Therapy Overview

Linear accelerator used to deliver radiation Deliver from multiple angles Fractionated treatment Therapeutic advantage

23

Treatment Planning Process

Take pre-treatment 4DCT scan Physician outlines target and critical structures this becomes problem data Planner solves optimization problem to produce treatment plan Go back and forth between planner and physician Traditionally, deliver same treatment every day over treatment course (little data re-acquisition and re-planning)
24

External beam radiation therapy

This is OK if the tumor doesnt move

25

Motion and motion uncertainty

Breathing motion trace

Purple outline = margin

26

Treatments trade off between tumour vs. healthy tissue

Aggressive treatment
(little motion uncertainty)

Balanced treatment
(moderate motion uncertainty)

Conservative treatment
(lots of motion uncertainty)

27

A Micro Problem: Summary

Use optimization to design radiation therapy treatments that
Target tumor Spare healthy organs Compensate for uncertain breathing motion

28

A Macro Problem: Clinic Scheduling

Hospitals schedule ambulatory clinics throughout the week Clinic schedule affects operations of many shared resources
Upstream blood lab Downstream chemo day care Nursing Rooms

Currently being studied at Princess Margaret Hospital and Womens College Hospital Goal: Create clinic schedule that balances shared resource utilization

29

Conclusion
Operations research has the capability to solve a wide range of practical problems, especially large-scale, complex, dataintensive ones Keys to success:
Access to reliable data and collaborators Formulating a good model (art vs. science) Recognizing limitations of model Translating abstract solutions into implementable recommendations

Operations researchers are always looking for challenges and collaborators in new fields
Wed like to believe we have a hammer for everybodys nail
30

This page intentionally left blank

31